Import AI 467: Self-sustaining AI viruses; pacing AI progress; confusion about AI and creativity

Researchers from the University of Toronto, Vector Institute, Cambridge, and ServiceNow have demonstrated a proof-of-concept computer virus that uses artificial intelligence to autonomously compromise machines and replicate itself. Per Import AI, the worm leverages open-weight language models running on stolen GPU resources to identify security vulnerabilities in its targets, devise tailored attacks, and spread to new hosts—all without relying on external vendor APIs. The attack succeeds in detecting vulnerabilities about eighty percent of the time and in self-replication roughly eighty-eight percent of the time, achieving an overall success rate near thirty-seven percent. The researchers built the agent with specialized reasoning nodes designed to focus its analysis on specific attack phases, and they note that the decentralized nature of the worm—operating as a swarm rather than a single point of failure—makes it particularly resilient to defensive measures.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton